This repository contains the data of all reviews of ICLR 2019 and ICLR 2020, crawled using the OpenReview API. See my blog for more information and plots.
Currently, only data for ICLR 2020 has been provided in a summary data format. See 2020.csv for more details.
The folders 'Data/2019' and 'Data/2020' contain all reviews of ICLR
2019 and 2020, respectively. Each folder follows the numerical ID that
OpenReview assigns to a paper. Each review is provided as a JSON
file.
-
Use the script
get_all_reviews.py
to download all the reviews of a given year. Currently, everything is hard-coded in the script. -
Use the script
extract_data.py
to extract data from the reviews and create aCSV
file. This is where most of the magic happens currently. I added a cursory calculation of NLP metrics but did not make use of the resulting values so far. -
Finally, to create the plots from the blog post, use the script
make_plots.py
.
I am actively working on all of these scripts, so stay tuned for improved versions.